CVE-2026-54044

LibreChat agent editors could redirect retained owner credentials

LibreChat-AI · LibreChat

First documented disclosure · Review checkpoint

Summary

An editor could change an action’s destination while preserving its owner’s secrets.

Severity and exploitation evidence

High · 7.7/10 · CVSS 3.1. Severity source.

Exploitation not established: Public evidence describes a weakness, source correction or controlled demonstration; malicious exploitation against users is not established.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

Action metadata merges retain API/OAuth credentials when domain/spec changes.

Conditions required

EDIT permission on a shared agent action with stored credentials and a later tool invocation by a permitted user.

Potential impact

Owner API keys or OAuth client secrets can be sent to the editor-controlled destination.

Affected versions and configurations

Before 0.8.7.

Fix and mitigation status

0.8.7.

Practical next steps

Upgrade to 0.8.7 or later and verify the deployed application image includes the correction.

Operating lesson

Invalidate retained credentials whenever their destination or authentication boundary changes.

Evidence and limits

Parent-agent authorization alone does not bind secrets to the action destination.0.8.7-rc1 also contains the fix. CNA publication 2026-11-20 is distinct from the earliest reviewed public source. GitHub_M assessment: CVSS 3.1 7.7.

Sources